wozaimoyu / acrnet Goto Github PK

This is a PyTorch implementation of ACRNet inference. The key results in paper Binarized Aggregated Network with Quantization: Flexible Deep Learning Deployment for CSI Feedback in Massive MIMO System can be reproduced.

The following requirements need to be installed.

• Python >= 3.7
• PyTorch == 1.6.0

As mentioned in the paper, our dataset is generated from COST2100 channel model. The preprocessed version provided by Chao-Kai Wen in CsiNet is recommended. You can download it from Google Drive or Baidu Netdisk.

Note that all dataset generation related setting can be found in the ACRNet paper if you wish to generate the dataset yourself.

The checkpoints of our proposed BCsiNet can be downloaded from Baidu Netdisk (passwd: he1o) or Google Drive

We recommend you to arrange the project tree as follows.

home
├── ACRNet  # The cloned BCsiNet repository
│   ├── dataset
│   ├── model
│   ├── utils
│   ├── main.py
├── COST2100  # COST2100 dataset downloaded following section A
│   ├── DATA_Htestin.mat
│   ├── ...
├── Experiments
│   ├── checkpoints  # The checkpoints folder downloaded following section B
│   │     ├── table1
│   │     ├── table2
│   ├── run.sh  # The bash script
...

The key results reported in Table I of the paper are listed as follows.

The key results reported in Table II of the paper are listed as follows. Note that all the compression ratio in Table II is 1/4.

In order to reproduce the aforementioned key results, you need to download the given dataset and checkpoints. Moreover, you should arrange your project tree as instructed. An example of Experiments/run.sh can be found as follows.

python /home/ACRNet/main.py \
  --data-dir '/home/COST2100' \
  --scenario 'in' \
  --pretrained '/home/Experiments/checkpoints/table1/cr4/1x_in/model.pth' \
  --batch-size 200 \
  --workers 0 \
  --reduction 4 \
  --expansion 1 \
  --gpu 0 \
  2>&1 | tee log.out

Note that the checkpoint must match exactly with the indoor/outdoor scenario and the hyper-parameter of ACRNet, including the reduction and the expansion. Otherwise the checkpoint loading would fail or the result will be incorrect.

This repository is modified from the BCsiNet open source code. Please refer to it if you are interested. The open source codes for CRNet and CsiNet can be helpful as well if you are interested in the benchmark networks.

Thank Chao-Kai Wen and Shi Jin group again for providing the pre-processed COST2100 dataset.